What the tweet revealed

Databricks wrote that Genie Agent Mode “investigates like a data analyst”. That framing matters because AI business intelligence tools are moving from natural-language SQL generation toward agentic analysis: planning an approach, checking hypotheses, running multiple queries, and explaining what the evidence supports.

The Databricks account is a first-party channel for platform releases across Lakehouse, AI/BI, Unity Catalog, and agent tooling. Its linked blog says Agent Mode enables users to ask a more advanced class of business questions: why something happened, what would happen under a different scenario, and how performance could improve. Instead of returning a single query, Genie can reason across several queries and produce a report.

How the mode works

The blog’s example starts with a spike in reopened support cases in December 2025. Agent Mode first confirms the spike, then explores possible drivers such as customers, products, categories, or teams. It uses business context from the Genie space, including Unity Catalog metadata and author-defined semantics, to focus on relevant factors. Databricks says the example reasoning trace executed 8 queries.

After testing hypotheses, Genie generates a report with findings, visualizations, and references to the underlying SQL. That last part is important for enterprise analytics: users need to inspect how an answer was produced before they trust it in a planning meeting. Databricks also says Agent Mode can dynamically scale reasoning to the question’s complexity, moving faster for simple prompts and spending more effort on deeper investigations.

The release is available through Workspace Previews, and AI/BI Dashboard users can use the experience from dashboards by default. The blog also points to future API support and unstructured document analysis, which would broaden Genie from dashboard Q&A into a larger enterprise agent surface.

What to watch next is whether Agent Mode improves accuracy on messy real warehouse schemas without hiding mistakes. The key test will be whether analysts can audit the SQL, correct semantics, and reuse reports rather than treating each answer as a one-off chat result. Source: Databricks source tweet · Databricks Genie Agent Mode blog

#ai-bi

Databricks Genie Agent Mode turns BI chat into analyst-style investigations